[GH-ISSUE #7232] Basic AI test result inconsistent compared to llama.cpp #51102

Open
opened 2026-04-28 18:22:53 -05:00 by GiteaMirror · 2 comments
Owner

Originally created by @brauliobo on GitHub (Oct 17, 2024).
Original GitHub issue: https://github.com/ollama/ollama/issues/7232

What is the issue?

For the following basic AI test:

I have 10 apples. I find 3 gold coins in the bottom of a river. The river runs near a big city that has something to do with what I can spend the coins on. I then lose 4 apples but gain a gold coin. Three birds run into my path and drop 6 apples each. I play an online game and win 6 gold coins but I have to share them equally with my 2 teammates. I buy apples for all the coins I have. The price of an apple is 0.5 coins. How many apples do I have? And where is the river?

With the correct result of 36 apples

I always get a different result when running with ollama compared to llama.cpp, see below.

ollama:

braulio @ phoenix ➜  ~  ollama run llama3.1-temp0                
>>> I have 10 apples. I find 3 gold coins in the bottom of a river. The river runs near a big city that has something to do with what I can spend the coins on. I then 
... lose 4 apples but gain a gold coin. Three birds run into my path and drop 6 apples each. I play an online game and win 6 gold coins but I have to share them equall
... y with my 2 teammates. I buy apples for all the coins I have. The price of an apple is 0.5 coins. How many apples do I have? And where is the river?
You started with 10 apples.

You found 3 gold coins in the river, but that doesn't affect your number of apples.

You lost 4 apples, so you had 10 - 4 = 6 apples left.

The birds dropped 6 x 3 = 18 apples. So now you have 6 + 18 = 24 apples.

You won 6 gold coins and shared them with your teammates, so you got 6 / 3 = 2 gold coins. You already had 3 gold coins from the river, plus the one you gained 
after losing an apple, for a total of 4 gold coins. Now you have 5 gold coins.

You can buy apples with your gold coins at a rate of 0.5 coins per apple. So you can buy 5 / 0.5 = 10 more apples.

Now you have 24 + 10 = 34 apples.

The river is near the city of Paris, France.

llama.cpp on the CPU

./llama-cli -m ../Meta-Llama-3.1-8B-Instruct-Q8_0.gguf --temp 0 -p "I have 10 apples. I find 3 gold coins in the bottom of a river. The river runs near a big city that has something to do with what I can spend the coins on. I then lose 4 apples but gain a gold coin. Three birds run into my path and drop 6 apples each. I play an online game and win 6 gold coins but I have to share them equally with my 2 teammates. I buy apples for all the coins I have. The price of an apple is 0.5 coins. How many apples do I have? And where is the river?"

# Step 1: Determine the initial number of apples and gold coins.                                                                                                      
I start with 10 apples and find 3 gold coins in the river.                                                                                                             
                                                                                                                                                                       
## Step 2: Calculate the change in the number of apples after losing some.                                                                                             
I lose 4 apples, so I have 10 - 4 = 6 apples left.                                                                                                                     
                                                                                   
## Step 3: Calculate the change in the number of gold coins after finding more.                                                                                        
I gain a gold coin, so I have 3 + 1 = 4 gold coins.
                                         
## Step 4: Calculate the number of apples dropped by the birds.                    
Three birds drop 6 apples each, so I gain 3 * 6 = 18 apples.                                                                                                           
                                                                                                                                                                       
## Step 5: Calculate the total number of apples after the birds drop some.         
I now have 6 + 18 = 24 apples.                                                                                                                                         
                                                                                                                                                                       
## Step 6: Calculate the number of gold coins won in the online game.                                                                                                  
I win 6 gold coins, but I have to share them equally with my 2 teammates, so I get 6 / 3 = 2 gold coins.
                                                                                                                                                                       
## Step 7: Calculate the total number of gold coins after the online game.         
I now have 4 + 2 = 6 gold coins.                                                   
                                                                                                                                                                       
## Step 8: Calculate the number of apples I can buy with the gold coins.                                                                                               
The price of an apple is 0.5 coins, so I can buy 6 / 0.5 = 12 apples.                                                                                                  
                                                                                   
## Step 9: Calculate the total number of apples I have after buying more.                                                                                              
I now have 24 + 12 = 36 apples.                                                                                                                                        
                                                                                                                                                                       
## Step 10: Determine the location of the river.
The river runs near a big city, but the problem does not specify which city.                                                                                           
                                         
The final answer is: $\boxed{36}$     

OS

Linux

GPU

Nvidia

CPU

AMD

Ollama version

0.3.12

Originally created by @brauliobo on GitHub (Oct 17, 2024). Original GitHub issue: https://github.com/ollama/ollama/issues/7232 ### What is the issue? For the following basic AI test: ``` I have 10 apples. I find 3 gold coins in the bottom of a river. The river runs near a big city that has something to do with what I can spend the coins on. I then lose 4 apples but gain a gold coin. Three birds run into my path and drop 6 apples each. I play an online game and win 6 gold coins but I have to share them equally with my 2 teammates. I buy apples for all the coins I have. The price of an apple is 0.5 coins. How many apples do I have? And where is the river? ``` With the correct result of 36 apples I always get a different result when running with ollama compared to llama.cpp, see below. ollama: ``` braulio @ phoenix ➜ ~ ollama run llama3.1-temp0 >>> I have 10 apples. I find 3 gold coins in the bottom of a river. The river runs near a big city that has something to do with what I can spend the coins on. I then ... lose 4 apples but gain a gold coin. Three birds run into my path and drop 6 apples each. I play an online game and win 6 gold coins but I have to share them equall ... y with my 2 teammates. I buy apples for all the coins I have. The price of an apple is 0.5 coins. How many apples do I have? And where is the river? You started with 10 apples. You found 3 gold coins in the river, but that doesn't affect your number of apples. You lost 4 apples, so you had 10 - 4 = 6 apples left. The birds dropped 6 x 3 = 18 apples. So now you have 6 + 18 = 24 apples. You won 6 gold coins and shared them with your teammates, so you got 6 / 3 = 2 gold coins. You already had 3 gold coins from the river, plus the one you gained after losing an apple, for a total of 4 gold coins. Now you have 5 gold coins. You can buy apples with your gold coins at a rate of 0.5 coins per apple. So you can buy 5 / 0.5 = 10 more apples. Now you have 24 + 10 = 34 apples. The river is near the city of Paris, France. ``` llama.cpp on the CPU ``` ./llama-cli -m ../Meta-Llama-3.1-8B-Instruct-Q8_0.gguf --temp 0 -p "I have 10 apples. I find 3 gold coins in the bottom of a river. The river runs near a big city that has something to do with what I can spend the coins on. I then lose 4 apples but gain a gold coin. Three birds run into my path and drop 6 apples each. I play an online game and win 6 gold coins but I have to share them equally with my 2 teammates. I buy apples for all the coins I have. The price of an apple is 0.5 coins. How many apples do I have? And where is the river?" # Step 1: Determine the initial number of apples and gold coins. I start with 10 apples and find 3 gold coins in the river. ## Step 2: Calculate the change in the number of apples after losing some. I lose 4 apples, so I have 10 - 4 = 6 apples left. ## Step 3: Calculate the change in the number of gold coins after finding more. I gain a gold coin, so I have 3 + 1 = 4 gold coins. ## Step 4: Calculate the number of apples dropped by the birds. Three birds drop 6 apples each, so I gain 3 * 6 = 18 apples. ## Step 5: Calculate the total number of apples after the birds drop some. I now have 6 + 18 = 24 apples. ## Step 6: Calculate the number of gold coins won in the online game. I win 6 gold coins, but I have to share them equally with my 2 teammates, so I get 6 / 3 = 2 gold coins. ## Step 7: Calculate the total number of gold coins after the online game. I now have 4 + 2 = 6 gold coins. ## Step 8: Calculate the number of apples I can buy with the gold coins. The price of an apple is 0.5 coins, so I can buy 6 / 0.5 = 12 apples. ## Step 9: Calculate the total number of apples I have after buying more. I now have 24 + 12 = 36 apples. ## Step 10: Determine the location of the river. The river runs near a big city, but the problem does not specify which city. The final answer is: $\boxed{36}$ ```` ### OS Linux ### GPU Nvidia ### CPU AMD ### Ollama version 0.3.12
GiteaMirror added the bug label 2026-04-28 18:22:53 -05:00
Author
Owner

@brauliobo commented on GitHub (Oct 17, 2024):

Reproduced the same inconsistency with llama3.2's ollama model and https://huggingface.co/bartowski/Llama-3.2-3B-Instruct-GGUF/blob/main/Llama-3.2-3B-Instruct-Q4_K_M.gguf with llama.cpp (tested on CPU and with CUDA, exactly the same result)

<!-- gh-comment-id:2418638405 --> @brauliobo commented on GitHub (Oct 17, 2024): Reproduced the same inconsistency with llama3.2's ollama model and https://huggingface.co/bartowski/Llama-3.2-3B-Instruct-GGUF/blob/main/Llama-3.2-3B-Instruct-Q4_K_M.gguf with llama.cpp (tested on CPU and with CUDA, exactly the same result)
Author
Owner

@brauliobo commented on GitHub (Nov 18, 2024):

is there a script to update llama.cpp source code? tried in different ways but it failed to compile. GGML_CALL for instance isn't in a header anymore...

<!-- gh-comment-id:2483691427 --> @brauliobo commented on GitHub (Nov 18, 2024): is there a script to update llama.cpp source code? tried in different ways but it failed to compile. GGML_CALL for instance isn't in a header anymore...
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: github-starred/ollama#51102